Giorgi / DuckDB.NET / 35847999295

88%

push

github

Keep ARRAY values aligned after a NULL array

ListVectorDataWriter placed ARRAY items at a running offset that only
advanced when an array was written. A NULL array never reaches
AppendCollection, so every array after it in the same batch was written
one row early, and the last row read uninitialised memory. This happened
wherever an ARRAY is involved: INTEGER[2], INTEGER[2][], INTEGER[][2]
and INTEGER[2][2].

An ARRAY stores no item positions of its own; DuckDB reads row r's items
at r * size. Derive the position from the row the same way. LISTs keep
the running offset, because each list entry records where its items
start.

A NULL array also has to mark its items NULL. DuckDB's appender hides
them only in its own copy of the validity, but for LIST items it reads
the list headers through the vector's own validity
(ListVector::GetConsecutiveChildListInfo), so an unmarked header under a
NULL array could hand it a garbage length. WriteNull is now virtual, and
ListVectorDataWriter overrides it to mark the items NULL, recursively for
nested arrays, as DuckDB's own Vector::SetValue does.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KxGp85FzaVDRPo7eXVBFrf

1598 of 1897 branches covered (84.24%)

Branch coverage included in aggregate %.

15 of 15 new or added lines in 1 file covered. (100.0%)

3178 of 3509 relevant lines covered (90.57%)

374909.76 hits per line